Does content length correlate with rankings and AI citations?

“Write more words” is some of the oldest advice in SEO, repeated so often it reads as settled fact. Two dated correlation studies, one aimed at classic Google rankings and one aimed at AI Overview citation, both put a number on that claim instead of just asserting it, and in both cases the number is smaller than the advice implies.

The AI-citation number: 0.04#

Ahrefs published a study on 3 December 2025 that pulled 560,346 AI Overviews, identified 1,677,876 cited URLs inside them, and successfully extracted usable content for 174,048 of those pages. Across that set, the Spearman correlation between a cited page’s word count and its citation was 0.04, a figure Ahrefs itself describes as essentially zero. The same dataset found 53.4% of cited pages run under 1,000 words, well short of the “comprehensive long-form guide” model that AI-search advice usually recommends. A correlation this close to zero means word count carries almost no predictive value for whether AI Overviews cite a page at all.

What the classic-ranking data shows#

Backlinko’s ranking-factor study, last updated 14 April 2025 and built on 11.8 million Google results, reports the average word count of a page-one result at 1,447 words. That number alone sounds like an endorsement of long content, but the same study found word count evenly distributed across positions 1 through 10: a page sitting at position 1 has no measurable word count advantage over a page sitting at position 10, once both have already reached the first page. Backlinko states directly that its design cannot explain why longer content tends to appear on page one in the first place, only that it does not appear to move a page’s position once it is there.

The shared confound both firms name#

Both studies point at the same alternative explanation for their own numbers: pages that cover a topic thoroughly tend to run long as a side effect, not because extra words themselves earn extra rank or citation weight. Ahrefs puts this plainly for its AI Overview data, arguing that long pages get cited because long content is simply what most publishers on a given topic produce, not because AI Overviews reward length on its own terms. Neither firm’s correlational design can separate “length causes visibility” from “topics that need depth produce length,” and both say so in their own published caveats rather than leaving readers to assume causation.

Why we care#

Treat word count as an output of doing the work, not a target to hit before publishing. A brief that specifies “cover these five subtopics with primary sources” will naturally land wherever it lands, sometimes 600 words, sometimes 2,500, and that length is a byproduct of what the topic required, not a lever pulled to satisfy Google or an AI Overview. Padding a thin brief to hit an arbitrary word count target does not move rankings by Backlinko’s own data, and does close to nothing for AI citation by Ahrefs’ own data. Write to the topic’s actual depth, then stop.